STMicroelectronics 2022
21
Binary Weighting
Quantizing data in a 1-bit (0, 1) form provides
two (2) great benefits in processing real-time
data with AI. First (#1), by simplifying data
from many bits to one, it reduces the need
for large memory. Second (#2), it simplifies
the computational complexity and efficiency
of the necessary calculations. This provides
faster computational results. It is recognized
that this simplification can possibly lose
precision. How is it that while precision is
reduced, excellent computational results
can still be achieved? Binary neural network
(BNN) is a network with binary weights and
activations that can be effectively trained
and run on systems with limited resources.
BNN is an extremely efficient form of deep
learning (DL). DL deals with a neural network
with more than two (2) layers and produces
automatic feature extraction (Figure 2).
Figure 2: An image of a neural network (NN) with six (6) layers.
(Source: Andrii/Stock.Adobe.com)
